Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77013.47401247967,
-      "required_pages_per_second_upper_estimate": 0.00532797082806254
+      "required_token_positions_per_second": 76886.8116712931,
+      "required_pages_per_second_upper_estimate": 0.00531920803340213
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9626.684251559958,
-      "required_pages_per_second_upper_estimate": 0.004030491668101323
+      "required_token_positions_per_second": 9610.851458911637,
+      "required_pages_per_second_upper_estimate": 0.004023862808445778
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19253.368503119917,
-      "required_pages_per_second_upper_estimate": 0.004009968913728047
+      "required_token_positions_per_second": 19221.702917823273,
+      "required_pages_per_second_upper_estimate": 0.004003373807388398
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28880.052754679873,
-      "required_pages_per_second_upper_estimate": 0.0017313150653558177
+      "required_token_positions_per_second": 28832.554376734908,
+      "required_pages_per_second_upper_estimate": 0.0017284676101238408
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28880.052754679873,
-      "required_pages_per_second_upper_estimate": 0.002245297071749229
+      "required_token_positions_per_second": 28832.554376734908,
+      "required_pages_per_second_upper_estimate": 0.0022416042817872926
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28880.052754679873,
-      "required_pages_per_second_upper_estimate": 0.37044230775233605
+      "required_token_positions_per_second": 28832.554376734908,
+      "required_pages_per_second_upper_estimate": 0.36983304955984286
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05319347438283728,
-    "producer_p95_seconds": 0.16769830500015814,
-    "gpu_update_seconds": 0.9573389714643016,
-    "producer_batches_per_second": 18.799298440311073,
-    "gpu_batches_per_second": 1.0445620932682247,
-    "producer_headroom": 17.997301033097852,
+    "producer_batch_seconds": 0.0706209557891313,
+    "producer_p95_seconds": 0.171616055999948,
+    "gpu_update_seconds": 0.958916079329734,
+    "producer_batches_per_second": 14.16010288767434,
+    "gpu_batches_per_second": 1.0428441253159326,
+    "producer_headroom": 13.578350343954323,
     "sustainable": true,
-    "buffer_seconds": 60.312355202251,
+    "buffer_seconds": 60.41171299777324,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 489298.1172003974,
-    "upload_bytes_per_second": 5419125.96073942,
+    "generation_bytes_per_second": 511520.35246432084,
+    "upload_bytes_per_second": 5346982.398202817,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.9919856739897,
+    "estimated_full_shards_per_hour": 109.76036005446643,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -460,6 +460,12 @@
       "bytes": 875600,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 0.8441450039999836,
+      "bytes": 578644,
+      "pages": 17,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -54,3 +54,4 @@
 {"step": 609800, "loss": 0.5783213973045349, "lr": 0.0001, "grad_norm": 0.16062904182535676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89257200749, "tokens_per_second": 153469.19244595084, "data_wait_s": 1.6522999430890195e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 609900, "loss": 0.5257395505905151, "lr": 0.0001, "grad_norm": 0.13698058650701322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89271990677, "tokens_per_second": 153815.20690466673, "data_wait_s": 1.3779000255453866e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 610000, "loss": 0.5316005945205688, "lr": 0.0001, "grad_norm": 0.14516027531668468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89286483603, "tokens_per_second": 150425.89357806955, "data_wait_s": 1.2999999853491317e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 610100, "loss": 0.5721548199653625, "lr": 0.0001, "grad_norm": 0.14133212756874464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89301011279, "tokens_per_second": 139268.55388615286, "data_wait_s": 1.18650004878873e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba02b8aba442ba0ecc88898f94038d770688872c276963bf473d19f237a84790
-size 120339
+oid sha256:aa83b7614675cc1b07e30e4afae69f504e827f557151bfc0d3ce19cf2d375aae
+size 122457